Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat(layered-storage): add a new storage for options #71

Draft
wants to merge 54 commits into
base: master
Choose a base branch
from

Conversation

Thomaash
Copy link
Member

@Thomaash Thomaash commented Dec 8, 2019

This is intended to replace the prototype plus hacks insanity used at the moment. However this is still WIP. There is no documentations yet and more testing is necessary.

TODO:

  • Write docs.
  • Add per key value validation.
  • Handle groups.
  • Discuss.
  • Do more testing.

In a quick test I was able to resolve the issue discussed in visjs/vis-network#178 and visjs/vis-network#213 with just a few lines of code. Which is much better than the massive mess of weird hacks that doesn't work reliably anyway. *

Putting this to use will be a lot of work but fortunately it should be possible to do it in parts. I would first use this in LayoutEngine and EdgesHandler to resolve the forementioned issues and then probably one module at the time.

Features:

  • Handles shorthand expansions.
  • Validates inputs via reusable functions.
  • Gives a feedback about invalid inputs. ***
  • Explicit layer/segment/key structure instead of prototype chains.
  • Observable. **
  • Encapsulates and automates options merging.
  • Higher layers override lower layers. *
  • Type safety in TypeScript.

* Hierarchical layout is incompatible with smooth edges and has to disable them. Overrides combined with observing easily and elegently solve that. See the forementioned issues for current state.
** This worked very nicely but it slowed exponentially with growing number of items rendering this feature pretty much unusable.
*** Now we sometimes ignore, sometimes log and sometimes throw. In other words it's an unpredictable mess. In LS a single function handles feedback for everything.

This is intended to replace the prototype plus hacks insanity used at
the moment. However this is still WIP. There is no documentations yet
and more testing is necessary.

TODO:
- Add off.
- Write docs.
- Probably some other things too.

In a quick test I was able to resolve the issue discussed in
visjs/vis-network#178 and visjs/vis-network#213 with just a few lines of
code. Which is much better than the massive mess of weird hacks that
doesn't work reliably anyway.

Putting this to use will be a lot of work but fortunately it should be
possible to do it in parts. I would first use this in LayoutEngine and
EdgesHandler to resolve the forementioned issues and then probably one
module at the time.

Features:
- Encapsulates options merging.
- Explicit layer/segment/key structure instead of prototype chains.
- Observable.
- Overrides. *
- Type safety in TypeScript.

* Hierarchical layout is incompatible with smooth edges and has to
disable them. Overrides combined with observing easily and elegently
solve that. See the forementioned issues for current state.
About half should be done by now.
It makes more sense given what it actually does.
Caching up front would be too complex to implement properly without
rebuilding the whole cache due to every change.
The sorting fails and TS already prohibits them. There's no indisputable
way of sorting nonnumbers anyway.
This is to simplify debugging. It logs the content of the storage to the
console using collased groups to denote the structure of the storage.
Previously the structures were kept even if all values were deleted.
BREAKING CHANGE: The on methods were removed from Layered Storage.

This was unbearably slow. Most of the time was actually spent handling
events. Making it faster without reducing it's functionality to the same
level as Emitter seems impossible to me. Since we already use the
Emitter there's no point in having such thing.
Prior to this all cache for given key was cleaned on set or delete.
However this is only necessary for monolithic segment. Other segments
can't affect anything but themselves so it's okay to keep the other
segments cached.

This greatly improves set and delete performance.
This greatly improves performance on mostly static data as it avoids
repeated traversing through the data just to find nothing over and over
again.
Before this all segments were iterated over and then checked for
existence. This is faster as it iterates only over existing cache
without checking for each segment's cache existence and also less code.

Also includes a test for possible regression that was unchecked before.
These are used to expand short hand values to their full forms.
BREAKING CHANGE: addValidators no longer exists

Appending validators to already existing one is most likely an error. At
least in Vis Network I see no reason to do it. Therefore it makes more
sense to throw in such a case.
BREAKING CHANGE: Some types were renamed.

It has the same format as the various .entries() methods so it makes
more sense to call it entries.
Those are a few micro optimalizations. However since this is executed
over and over again it adds up to noticable improvement.
BREAKING CHANGE: Most stuff was changed.

This is a really big overhaul that greatly reduces the API and the
codebase. This should be much easier to maintain and also more
performant when used at the expense of removing support for some niche
features.
This will make integrating this into existing codebase much easier.
We'll be able to mostly contain the changes into .setOptions() methods.
Prior to this only the commit would fail.
This can be used to implement groups from Vis Network. It actually
allows for multiple groups per single node which is already a wanted
feature that would be extremely difficult to implement using the
(anti)pattern that's currently in Vis Network for options.
This also includes a tweak to the way validating is done.
BREAKING CHANGE: The types are different, though the code is pretty much
the same on the outside.

This allows for typesafe expanding of values or transforming them
(expanders allow both at the same time).
BREAKING CHANGE: This completely prevents null from being used as a
value.

We could change it to a symbol if this turns out to be a problem.
Since it's not possible to use custom tsconfig.json it only works with
what the authors configured it to work with. There's an issue on their
repo tracking this.

It doesn't work with layered storage because it uses iterables other
than arrays and that's not configured in check-dts.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

1 participant